Benchmark proposed and used in: Touch-based Curiosity for Sparse-Reward Tasks.
**MiniTouch**, a manipulation benchmark of simulated tasks is comprised of four manipulation tasks 1)pushing 2)opening 3)picking 4)playing with simple objects. It allows evaluation of models' performance on different manipulation tasks that can leverage cross-modal learning.Clone the repository then:
cd minitouch
pip install -e .
import minitouch.env
import gym
env = gym.make("Pushing-v0")
state = env.reset()
for i in range(0, 1000):
state, reward, done, info = env.step(env.action_space.sample())
- Pushing-v0
- Opening-v0
- Picking-v0
- Playing-v0
Note: If you want to see the GUI of the environment you have to use the debug version of the environment. For example use "PushingDebug-v0" instead of "Pushing-v0".